Notice of Special Interest (NOSI): Explainable Artificial Intelligence for Decoding and Modulating Neural Circuit Activity Linked to Behavior

Funding Agency:
National Institutes of Health

The eXplainable Artificial Intelligence (XAI) framework aims to provide strong predictive value along with a mechanistic understanding of AI solutions by combining machine learning techniques with effective explanatory techniques. This Notice of Special Interest (NOSI) solicits applications in the area of XAI applied to neuroscientific questions of encoding, decoding, and modulation of neural circuits linked to behavior. This NOSI encourages collaborations between computationally and experimentally focused investigators. This NOSI seeks the development of machine learning algorithms that are able to mechanistically explain how experimental manipulations affect cognitive, affective, or social processing in humans or animals. Proof-of-concept applications aimed at improving the current state of the technology that uses XAI to provide unbiased, hierarchical explanations of causal relationships between complex neural and behavioral data are also appropriate.

Despite the rapid growth and adoption of machine learning and artificial intelligence (AI) techniques to scientific questions, the lack of insight into the inner workings of these approaches has impeded full scientific understanding that leads to machine-identified neuro-behavioral mechanisms. However, machine learning techniques have often been applied to categorize and predict neural and behavioral outcomes without providing a mechanistic understanding of what drives those predictions and classifications. Understanding the mechnistic factors critical to a machine-learning-based outcomes may lead to the identification of novel neurobehavioral solutions, theories, and potential targets for further studies or for intervention development.XAI consists of artificial intelligence algorithms in which the processes of arriving at final actions (e.g., predictions, classifications, and recommendations) can be easily understood by its users. XAI aims to overcome limitations of classical machine learning, including a lack of transparency and non-generalizability, by keeping the human-in-the-loop. While optimizing for accuracy or performance, a standard AI may learn useful rules from the specific training set. However, it may also learn inappropriate or non-generalizable rules. XAI provides methods to examine existing machine learning models more closely and new approaches that are explicitly designed to provide greater transparency. In a transparent XAI framework, users will have the ability to audit specific machine-identified rules/hypotheses and to discover how how much of the outcome variance those rules explain and how likely it is that the system will generalize outside a specific training set.

XAI is about enhancing machine-human collaborative intelligence in a new model in which researchers and end-users co-work with AI systems rather than using them as tools. As in most successful collaborations, each brings to the table abilities that the other lacks. NIMH promotes a deep mechanistic understanding of normative and abnormal neurobehavioral brain functions linked to mental health and the pathophysiology of psychiatric disorders. NIMH is interested in transforming classical ‘black box’ machine learning models into XAI ‘glass box’ models, without significantly sacrificing performance. The goal of this NOSI is to encourage investigators to apply XAI techniques to further our understanding of the neural circuitry linked to behavior and to improve our understanding of therapeutic strategies to enhance cognitive, affective, or social function. To develop new treatments for mental illness, a better understanding of how to modulate neural dynamics responsible for complex functional domains and/or maladaptive behaviors is critical. In order to achieve this understanding using XAI techniques, collaborations between computational and experimental investigators are strongly encouraged. In the context of mental health, the amount and type of explanatory information accessed may vary based on the stakeholder (clinicians, patients, or researchers) interacting with the AI system. Projects developing XAI for use in animal and/or human research are appropriate to this announcement. Human studies may involve healthy controls, community samples, and/or patient populations.

The Office of Research on Women’s Health focuses on research that is relevant to the health of women across the life course and advancing science where the consideration of sex and/or gender influences on health are integrated across the biomedical research enterprise, as highlighted in the 2019-2023 Trans-NIH Strategic Plan for Women's Health Research. Computational Psychiatry uses modeling tools, integrating multiple levels and types of analysis, to enhance understanding and treatment of psychiatric illness and prediction of behavior/symptom change. In the context of this FOA, ORWH is interested in supporting studies where principles of computational modeling are employed to explore sex and/or gender differences and/or health disparities questions relevant to psychopathology. Advancing rigorous and ethical research to understand the fundamental relationship between sex and gender-specific symptoms and underlying neurobiological function leading to clinically useful applications/intervention insights for populations of women that bear a disproportionate burden of risks and poorer outcomes are of particular interest.

  • The design and testing of computational models to imitate sex and gender differences in clinical phenotypes enabling investigation of underlying neurobiological function that correlates with psychiatric symptoms/somatic responses
  • To design and test simulation modeling tools for psychiatric symptoms to enable study of emotional, behavioral and physiological responses differences in psychiatric disorders by unique population level psychosocial risk factors (e.g. social determinants frequently associated with poor health in marginalized communities)

This notice applies to due dates on or after February 5, 2023 and subsequent receipt dates through February 5, 2026.





Engineering and Physical Sciences
Medical - Basic Science
Medical - Clinical Science
Medical - Translational

External Deadline

June 5, 2024